Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Aeronautical and Aviation Engineering | en_US |
| dc.contributor.advisor | Wen, Weisong (AAE) | en_US |
| dc.contributor.advisor | Hsu, Li-ta (AAE) | en_US |
| dc.creator | Zheng, Xi | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/14373 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Accurate and safety-quantifiable localization methods for autonomous robotic systems | en_US |
| dcterms.abstract | Accurate and safety-quantifiable localization is critical for safety-critical autonomous robotic systems, such as unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UAVs). While visual-inertial odometry (VIO) provides high-frequency pose estimation, it inherently suffers from cumulative drift. Meanwhile, current pose estimation approaches for localization predominantly rely on local nonlinear optimization techniques, which lack guarantees of global optimality and are vulnerable to outliers. Additionally, a safety-critical robotic system is required not only to achieve accurate pose estimation but also to predict and quantify the associated localization uncertainty by providing an error bound for each pose. These limitations severely compromise the accuracy, reliability, and safety of localization, despite its pivotal role in safety-critical applications. | en_US |
| dcterms.abstract | To address these challenges, this thesis develops a comprehensive three-stage framework for robust and safety-quantifiable localization. The first study introduces a novel line feature-based visual localization method that leverages prior 3D maps while incorporating safety quantification strategy. Building upon VIO for high-frequency local pose initialization, the proposed method establishes geometric constraints between 2D image lines and 3D lines belonging prior map through an innovative foot-point based association. The safety quantification mechanism, inspired by global navigation satellite systems (GNSS) receiver autonomous integrity monitoring (RAIM) methodology, employs a weighted sum of squares residual with Chi-squared testing for outlier rejection, complemented by a derived protection level scheme that bounds potential errors in both position and orientation domains. | en_US |
| dcterms.abstract | The second study advances the first work through a tightly-coupled factor graph formulation that fully integrates 3D prior line maps with VIO measurements, addressing the complementarity limitations of loosely-coupled approaches. This framework incorporates a robust cross-modality line association model between 3D maps and 2D images, enhanced by an efficient line-tracking strategy for outlier rejection. A theoretically grounded line feature cost model is developed for factor graph optimization, supported by the first rigorous observability analysis of such systems. The analysis reveals full observability of global translations with only the yaw angle around gravity remaining unobservable, providing critical insights for safety-critical applications. | en_US |
| dcterms.abstract | The final study overcomes fundamental limitations of local optimization methods by adopting a global estimation framework that unifies outlier rejection with estimation error prediction. The solution employs Truncated Least Squares (TLS) for robust estimation, with the non-convexity challenge addressed through a Graduated Non-Convexity (GNC) strategy combined with Gaussian-Newton optimization. This approach systematically approximates and solves progressively more challenging subproblems to converge to globally optimal solutions. Furthermore, the framework incorporates an error propagation model that explicitly accounts for measurement uncertainties and residual outliers, enabling predictive error bounding to guarantee estimation safety throughout the localization process. | en_US |
| dcterms.abstract | This integrated three-stage approach systematically addresses the accuracy, robustness, and safety quantification challenges in modern localization systems, while providing theoretical guarantees and practical implementation strategies suitable for safety-critical robotic applications. The feasibility and practicality of the proposed frameworks have been validated through simulations and real-world datasets. To benefit the research community, we have open-sourced the complete implementation, including algorithms and datasets. | en_US |
| dcterms.extent | xxi, 159 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2026 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14373